Re: [nemo] RE: draft-ietf-nemo-home-network-models-05

[Alexandru Petrescu <[e-mail]>]

Pascal Thubert \(pthubert\) wrote:
> Hi Alex
> 
>>> In fact, the text did not appear particularly useful, and would 
>>> have lead to discussion going above and beyond the scope of the 
>>> draft; so the text was simply removed.
>> 
>> If that ICMP Redirect text (between HA and MR) did not appear to be
>>  particularly useful then I wonder how proxy ND by HA for _all_ 
>> addresses in the moving network is of any use at all.  By RFC3963 
>> the HA does proxy ND only for MR's Home Address.
> 
> [Pascal] We removed it because some nodes (including routers) will 
> not react to the redirect, see 
> http://www.mobilenetworks.org/~pthubert/draft-ietf-nemo-home-network-mod
>  els-issue9.txt so we can not see that as an absolute solution, 
> making it hard to recommend. Routing Information will not affect the
>  Hosts/MNs on the Home Links, so it is also a problem. We end up with
>  a pure L2 solution, and since the MR is reachable over a tunnel, it
>  boils down to proxying techniques.

Thanks for documenting that issue.

Pascal:
> ICMP redirect is a router to host thing.

Right, and the Mobile Router acts as a Mobile Host too.

Ryuji:
> some nodes ignore ICMP redirect.

Right and other nodes will never do proxy ND for addresses that are
topologically incorrect on that link.

>> draft-ietf-nemo-home-network-models-xx.txt:
>>> Thus, on the Home Link, the Home Agent must intercept all the 
>>> packets to ALL the Mobile Network Nodes on the registered 
>>> prefixes.  In order to do so, the Home Agent might perform some 
>>> form of ND proxying for all addresses in all registered Mobile 
>>> Network Prefixes.
>> 
>> This above paragraph makes no particular sense at all, since a HA 
>> doing proxy ND for a LFN node which is below MR will intercept 
>> packets from all the neighbours of HA addressed to LFN, even though
>>  it is the MR who should intercept or receive those packets.
>> 
>> There can be no "proxy ND" for an LFN on HA's link as long as the 
>> normal ND of same LFN is not happening on the home link.
> 
> [Pascal] Please read again. The "Nodes on the registered prefixes" 
> means nodes  (eg. LFNs) attached to MRs that are away from home. Thus
>  MRs can not intercept the packets to the LFN over the Home Link.

For some reason "intercept" was perceived by you as something where MR
must do proxy ND in order to receive.  I didn't mean so, sorry.  I meant 
MR should "receive" (or "intercept") packets for LFNs when away from 
home.  This happens by MR receiving packets tunnelled by HA.  The HA has
"intercepted" those packets for LFNs by doing proxy ND for MR's Home
Address exclusively, and not by doing proxy ND for any LFN address.  It
makes no sense for HA to do proxy ND for LFN address.

Again, HA doing proxy ND for LFNs is something topologically invalid.
Proxy ND of a host should be done on a link where the real ND for that
host is topologically valid.

Moreover, there are potentially 2^64 LFNs behind an MR with MNP /64.  HA
has no means to know which of those are active.  So will it send 2^64
periodic NAs as part of its proxy ND for LFNs?

> If MR was at home, then section 6.2.1 or 6.2.2 would apply. 6.2.1 
> describes your point, and yes, in that case the MR does the 
> interception, not the HA.

We agree on this.

Pascal:
> Thus, on the Home Link, the Home Agent must intercept all the packets
>  for ALL the Mobile Network Nodes on the registered prefixes - that 
> is for ALL nodes attached to Mobile Routers that are away from Home. 
> This should be a layer 2 operation, rather than layer 3.  The Home 
> agent might, for example,  perform some form of ND proxying for all 
> addresses in all registered Mobile Network Prefixes.  The Home Agent 
> must also protect the MNP space from autoconfiguration by 
> uncontrolled visitors at Neighbor Discovery level.

Is there a known implementation where the HA intercepts all the packets 
for ALL the Mobile Network Nodes by doing proxy ND for their addresses?

Does anyone plan to write such an implementation?

Alex